Deep sequencing of the TCR‐β repertoire of human forkhead box protein 3 (FoxP3)⁺ and FoxP3^– T cells suggests that they are completely distinct and non‐overlapping

Summary

Maintenance of peripheral tolerance requires a balance between autoreactive conventional T cells (T_conv) and thymically derived forkhead box protein 3 (FoxP3)⁺ regulatory T cells (tT_regs). Considerable controversy exists regarding the similarities/differences in T cell receptor (TCR) repertoires expressed by T_conv and tT_regs. We generated highly purified populations of human adult and cord blood T_conv and tT_regs based on the differential expression of CD25 and CD127. The purity of the sorted populations was validated by intracellular staining for FoxP3 and Helios. We also purified an overlap group of CD4 T cells from adult donors to ensure that considerable numbers of shared clonotypes could be detected when present. We used deep sequencing of entire TCR‐β CDR3 sequences to analyse the TCR repertoire of T_conv and tT_regs. Our studies suggest that both neonatal and adult human T_conv and tT_reg cells are, in fact, entirely distinct CD4 T cell lineages.

Introduction

Homeostatic tolerance versus autoimmunity is determined, in part, by the ratio between autoreactive conventional T cells (T_conv) and thymically derived regulatory T cells (tT_regs) specific for self‐antigens. It has been shown previously that tT_regs develop separately from T_conv in the thymus 1, 2, contain a unique epigenetic landscape shaping their distinct gene expression profiles during thymic development 3 and continue to be marked by unique cell‐surface and intracellular proteins in the periphery 4. Despite these differences, both tT_regs and T_conv have the potential to respond to an identical array of peripheral self‐antigens in a particular end‐organ niche 5. Previous studies of T cell receptor (TCR) repertoires in both mice and humans have estimated the overlap between tT_regs and T_conv to range from as low as 3 to 30%. These studies have either relied upon studying rearrangement on a fixed Vα background and limited Vβ recombination partners (mouse) 6 or have focused upon a limited set of endogenous Vβ gene members (humans) 7. In addition, isolation of pure populations of tT_regs in humans for repertoire studies previously relied primarily upon expression of high levels of CD25 7, 8. A further complication in comparison of TCR repertoires is that some T_regs might be generated peripherally from T_conv cells (pT_regs for ‘peripherally’ generated), and these pT_regs would share the same repertoire as T_conv cells 9.

In this study, we have re‐explored the question of TCR repertoire differences between T_conv and tT_reg by performing comprehensive deep sequencing of all TCRVB genes. We have also prepared highly purified tT_regs (CD25^hi and CD127^low) and T_conv (CD25^–CD127⁺) and validated the purity of the preparations from either cord blood or adult peripheral blood mononuclear cells (PBMCs) by intracellular staining for forkhead box protein 3 (FoxP3) and Helios. A third population (CD25^–CD127^–) was also isolated from adult samples, and this population contained both FoxP3⁺Helios⁺ tT_reg cells and FoxP3^–Helios^– T_conv. This ‘overlap’ group allowed us to confirm that our protocol for TCR deep sequencing is sensitive enough to detect shared clonotypes between isolated CD4⁺ T cell subsets when true overlap exists. We have found that tT_reg and T_conv TCR repertoires are almost completely distinct and non‐overlapping. The implications of these results for understanding the separate development of tT_reg and T_conv lineages are discussed.

Materials and methods

Healthy individual blood samples

Institutional Review Board‐approved buffy coats were obtained from the NIH blood bank donated by healthy volunteers. Cord blood samples were purchased from the New York Blood Center.

Isolation of cells for RNA purification

PBMCs were prepared by Ficoll centrifugation. CD4⁺ T cells were enriched by negative isolation using an AutoMacs Pro (Miltenyi Biotec, San Diego, CA, USA). CD25 and CD127 staining was followed by cell sorting for three populations. T_conv, tT_regs and overlap groups were validated by Helios/FoxP3 staining. Purified RNA was prepared using Trizol (or equivalent) of 2 million cells from each cell population (adults) or, from cord blood, 1 million sorted T_conv and 70 000 sorted tT_reg.

TCR‐β deep sequencing and analysis

Sample preparation, library construction and sequencing were performed as described previously 10. TCR‐β annotation was performed by combining a custom Java program written in house and the National Center for Biotechnology Information's blast program. Briefly, TCR‐β V and TCR‐β J germline genes of a TCR‐β read were identified first, and the CDR3 was determined by finding the conserved cysteine and phenylalanine at 5′ and 3′ ends of the CDR, respectively. For each combination, the number of nucleotides contributed by germline V‐, D‐ and J‐ genes were calculated.

Shannon entropy, species richness and evenness were calculated for each TCR‐β repertoire using the R package, Vegan (http://r-forge.r-project.org/projects/vegan/). The entropy and richness were normalized by calculating the maximum Shannon entropy and maximum combination richness for each repertoire based on the cell count used for library generation. The calculated Shannon entropy and combination richness were then divided by the respective maximums to return a number between 0·0 and 1·0. For each repertoire, the average and standard deviation (s.d.) were calculated for germline index and CDR3 length. Custom Perl scripts were used to calculate the distribution of CDR3 length, V–J pairing percentage and amino acid compositions of each CDR3 position, as well as whole CDR3s from all the annotated TCR‐β sequence reads in each subject and group. Regarding the sharing, the Venn diagrams are overlaps of clonotypes (not taking expansion into account). The heat‐maps are the sum of the frequencies of the shared clones (which takes expansion into account). All the statistical analyses were performed with GraphPad Prism version 6 (GraphPad Software, La Jolla, CA, USA) and computing environment r software; additional packages (ggplot2 and heatmap.2) were taken from the Comprehensive R Archive Network (for additional details see reference [10]).

For comparison of the CDR3 length distribution between adult donor CD4 T cell subgroups, the Kolmogorov–Smirnov test was applied to the 25 most expressed V genes comparing T_conv, T_reg and overlap groups for adult samples 118 and 210 and generated a table of P‐values and D‐statistics. A two‐sample Kolmogorov–Smirnov test (K–S test) determines whether the distributions of some parameter (in our case, CDR3 length) for two different samples appear to be sampled from the same population (null hypothesis) or if the distributions indicate that they are actually two separate populations (alternative hypothesis). The D‐statistic indicates the largest difference between the two populations being compared and the P‐value is the significance of that largest difference. See https://en.wikipedia.org/wiki/Kolmogorov%E2%80%93Smirnov_test.

Results

Isolation of tT_regs, T_conv and an overlap group

The cell sorting strategy used to isolate tT_regs, T_conv cells and the overlap group from adult samples based on the differential expression of CD25 and CD127 is outlined in Fig. 1a. As reported previously, CD25^hiCD127^low cells are highly enriched for FoxP3⁺Helios⁺ tT_regs (≥ 90%, Fig. 1b), whereas CD25^–CD127⁺ cells are enriched (≥ 95%) for FoxP3^–Helios^– T_conv. The isolated tT_reg group contained a small number (∼5%) of FoxP3^–Helios⁺ cells and FoxP3⁺Helios^– cells (Fig. 1b). The FoxP3^–Helios⁺ cells are likely to be T_conv (our unpublished observations) that would ‘contaminate’ the purified tT_regs and potentially increase the likelihood of shared sequences between tT_reg and T_conv (CD25^–CD127⁺) cells. The FoxP3⁺Helios^‐ population has been shown by our laboratory to contain both T_reg (including possibly tT_reg and pT_reg) and T_conv and would therefore also potentially increase the likelihood of shared sequences with the T_conv group (CD25^–CD127⁺).

Figure 1.

Sorting strategy for enriching three different populations of CD4 T cells. (a) A representative flow cytometry histogram is shown for CD127 versus CD25 staining on gated CD4 T cells prior to and subsequent to sorting for the conventional T cell (T_conv) population (CD4⁺CD25^–CD127⁺), thymically derived regulatory T cell (tT_reg) population (CD4⁺CD25⁺CD127^–) and overlap population containing both T_conv and tT_regs (CD4⁺CD25^–CD127^–). (b) Helios and forkhead box P3 (FoxP3) intracellular staining of CD4 T cells prior to and subsequent to sorting based on surface markers as shown in (a) indicates the purity of each sorted population. Effector T cells (T_eff) are used interchangeably with conventional T cells (T_conv).

The overlap group (CD25^–CD127^–) contains both FoxP3⁺Helios⁺ tT_regs (5–8%) as well as FoxP3^–Helios^– T_conv (80–90%) and, therefore, should be expected to contain clonotypes that should be shared with both tT_reg and T_conv groups. If sequenced tT_reg and T_conv group clonotypes showed minimal overlap, a potential criticism would be that this was due to sampling error and that potential shared clonotypes were not detected due to limited coverage of all potential sequences. We included the overlap group precisely to show that if shared clonotypes could be detected, i.e. between the overlap group and either the tT_reg or T_conv group, this would strengthen the conclusions from the comparison of the tT_reg and T_conv groups to each other from an individual donor.

Isolation and amplification of TCR‐β RNA, followed by deep sequencing and bioinformatics analysis, yields a significant number of unique CDR3 clonotypes

In order to achieve good coverage of potential TCR‐β clonotypes, 2 million cells were isolated for each group from two separate adult individuals. Richness represents the numbers of unique TCRs in a given sample. The richness index was calculated by dividing the measured richness by the maximum richness (i.e. the number of cells used to generate a library, where it is assumed that each cell could possibly contain a unique TCR). Evenness describes how evenly distributed the frequencies of each unique TCR are across the population measured. Shannon entropy is a measure of diversity that takes both richness and evenness into account. The Shannon entropy index was determined by dividing the measured Shannon entropy by the theoretical maximum Shannon entropy that is determined by assuming maximum evenness, and each cell used to generate a sequencing library could possibly contain a unique TCR 10, 11.

Sequencing was not shallow. For the adult samples they were sequenced, on average, to a coverage (annotated reads/cells used to generate the library) of 1·51 (s.d. = 0·65). For the cord blood samples they are sequenced, on average, to a coverage of 1·66 (s.d. = 0·44). This means that, for all samples, the TCR‐β transcript was sampled more than once.

Deep sequencing yielded 2700–6900 high‐confidence individual CDR3 sequences for tT_reg, T_conv and overlap groups from two separate healthy adult individuals. The normalized Shannon Index ranged from 0·39 to 0·58 and richness ranged from 0·001 to 0·003 10.

Analysis of adult TCR‐Β CDR3 clonotypes shows little to no overlap between tT_regs and T_conv

A number of algorithms were used to analyse the CDR3 clonotypes. Venn diagrams indicate the total number of unique clonotypes as well as the number of shared clonotypes between tT_regs, T_conv and overlap groups (Fig. 2, top). In order to determine overall sharing, it is necessary to take into account the abundance of each clone. Theoretically, even a limited number of shared clonotypes may contribute to overall significant overlap if these shared clonotypes are represented more highly in the populations. Regarding the sharing, the Venn diagrams are overlaps of clonotypes (not taking expansion into account). The heat‐maps are the sum of the frequencies of the shared clones (which takes expansion into account). In the earlier study of CDR3 sharing, it was estimated that the percentage of shared unique clonotypes was approximately 3% between tT_regs and T_conv; however, due to high abundance of these clonotypes, the overall overlap was calculated as 24% 7. This study concluded that tT_regs and T_conv do not share many TCR‐β CDR3 sequences, but when they do, these correspond to highly frequent T clonotypes. However, in our study, which included significantly greater coverage of all TCRBV genes, not only did we find minimal sharing of individual clonotypes, but we also found less than 1% overall sharing of clonotypes (Fig. 2, bottom panels). This indicates that there are a few shared clonotypes, but that these are not high frequency clonotypes. The difference between our findings and the previous study are due probably to the fact that only one or two TCRBV–J pairings were used per donor 7, whereas we looked at more than 200 TCRBV–J pairings per donor.

Figure 2.

Minimal overlap of T cell receptor (TCR) VB CDR3 sequences between adult thymically derived regulatory T cell (tT_reg) and conventional T cell (T_conv) populations. Two healthy adult donors (118 and 210) were analysed for sharing between tT_reg, T_conv and overlap groups, as defined in Fig. 1. The top Venn diagrams indicate the numbers of unique CDR3 clonotypes per group and indicate the number of shared clonotypes between groups. The lower 3 × 3 tables take into account CDR3 clone abundance and indicate the fraction (indicated in individual squares) of CDR3 clonotypes from a particular group (y‐axis) that is shared with a different group (x‐axis).

tT_reg and T_conv repertoires from cord blood are similarly diverse, but direct sharing of CDR3 clonotypes is less than 1%

In order to determine if tT_reg and T_conv TCR repertoires are already distinct early after neonatal thymic development, we performed the same purification strategy as above on two cord blood samples. In contrast to adult peripheral blood, cord blood contained few to no cells in the overlap group of CD25^–CD127^– CD4 T cells. This is due probably to the fact that this group is enriched in memory cells and/or peripherally converted T_regs (data not shown). Deep sequencing yielded 6000–46 000 high‐confidence individual CDR3 sequences for tT_reg and T_conv groups from two separate healthy cord blood donors. The normalized Shannon index ranged from 0·82 to 0·87; richness index ranged from 0·005 to 0·11 10. Venn diagrams shown in Fig. 3 indicate the total number of unique clonotypes as well as the number of shared clonotypes between tT_regs and T_conv groups (Fig. 3, top). Heat‐maps shown in Fig. 3 take into account both number and abundance of shared clones. The data again indicate that overall sharing was minimal between cord blood tT_regs and T_conv cells.

Figure 3.

Minimal overlap of T cell receptor (TCR) VB CDR3 sequences between cord blood thymically derived regulatory T cells (tT_regs) and T_conv cells. Two healthy cord blood donors (6 and 7) were analysed for sharing between tT_reg and T_conv groups, as defined in Fig. 1. The Venn diagrams indicate the numbers of unique CDR3 clonotypes per group and indicate the number of shared clonotypes between groups. The lower 3 × 3 tables take into account CDR3 clone frequency and indicate the fraction (indicated in individual squares) of CDR3 clonotypes from a particular group (y‐axis) that is shared with a different group (x‐axis).

Analysis of V–J pairing indicates broad TRBV and TRBJ usage in tT_regs and T_conv in both adult peripheral blood as well as cord blood

Heat‐map TCRBV–J grids show that tT_regs and T_conv from two separate adult donors use a very broad number of TRBV and TRBJ gene members (Fig. 4,b). Certain TRBV and TRBJ genes are very highly used. For example, TRBV5‐1, TRBV20‐1 and TRBV20‐1 paired with TRBJ2‐1 through TRBJ2‐3 represent a significantly high number of tT_reg clonotypes from both adult donors (Fig. 4), as well as from both cord blood donors (Fig. 5). Whereas overall frequencies of TRBV/TRBJ pairings differ slightly for adult tT_regs compared to T_conv cells (Fig. 4), there is great similarity between frequencies of paired V and J segments for cord blood tT_regs and T_conv cells (Fig. 5). The number of total samples (two adults and two cord blood donors) was not sufficient to perform statistical analysis on V–J pairing similarities or differences between tT_regs and T_conv cells.

Figure 4.

Specific V–J pairings between thymically derived regulatory T cell (tT_reg) and conventional T cell (T_conv) populations from two adult donors. TRBV genes paired with TRBJ genes are shown with relative frequency among unique clonotypes for adult donor 118 (a) and 120 (b). The colour key indicates the frequency of individual V–J pairings.

Figure 5.

Specific V–J pairings between thymically derived regulatory T cell (tT_reg) and conventional T cell (T_conv) populations from two cord blood donors. TRBV genes paired with TRBJ genes are shown with relative frequency among unique clonotypes (colour code) for cord blood donor 6 (a) and cord blood donor 7 (b). The colour key indicates the frequency of individual V–J pairings.

Potential error in focusing analysis on a limited set of V(D)J genes

Previous analysis of human T_reg TCR repertoire often relied upon polymerase chain reaction (PCR) amplification of a limited set of TRBV genes, followed by extrapolation to the entire repertoire 7. To demonstrate how this approach might lead potentially to erroneous interpretation, we chose a relatively abundant pairing of V, D and J genes to investigate overlap between tT_regs and T_conv cells. As shown in Fig. 6a the TRBV20‐1_TRBD2_TRBJ2‐3 was one of the most frequently used V(D)J recombination for both adult donors. Comparing unique TRBV20‐1_TRBD2_TRBJ2‐3 clones, we observed only limited sharing of clones between adult tT_regs and T_conv cells (Fig. 6b). Interestingly, one of the shared TRBV20‐1_TRBD2_TRBJ2‐3 clones in adult 118 tT_regs was highly abundant, resulting in a calculation of > 50% of tT_reg‐specific total clones being shared with T_conv (Fig. 6c). This highlights how focusing the analysis on only one or two TRVB–VJ pairings could potentially skew the data to suggest much greater sequence overlap than is present in the overall TCR repertoires.

Figure 6.

Overlap of clones using a commonly used TRBV/TRBJ pair in adult donors. (a) One of the most highly used TRBV genes for the two adult donors was TRBV20‐1, which was paired frequently with TRBJ2‐3. (b) Venn diagrams and (c) 2 × 2 tables indicate the overall sharing of unique clones and total sharing of clones, respectively, encoded specifically by the recombination of TRBV20‐1/TRBD2/TRBJ2‐3.

We conducted a similar analysis for the 16 most abundant TRBV genes in one of the cord blood donors (Fig. 7). The table shown in Fig. 7b indicates the specific TRBV gene, for which the corresponding CDR3 amino acid length distributions are shown in Fig. 7b. For many of the TRBV genes there appear to be similar, although not identical, CDR3 length distributions (Fig. 7a). However, as shown in Fig. 7b, the frequency of shared clones ranges from zero for the majority of TRBV genes to as high as 1·6% shared clones by DNA sequence and 4·6% shared clones by amino acid sequence for TRBV5‐1. This analysis again shows how focusing upon a limited number of TRBV genes can result in a skewed interpretation of overlap between tT_reg and T_conv repertoires.

Figure 7.

Length distribution and individual overlap of DNA and amino acid sequences for the most abundant TRBV genes in one cord blood donor. CDR3 amino acid length distributions are shown for conventional T cells (T_conv) and thymically derived regulatory T cells (tT_regs) (a) for the 16 most abundant TRBV genes. The tables shown in (b) serves as a key for which TRBV gene is represented in each CDR3 length distribution. Also shown in (b) are the % of tT_reg clones that are shared with T_conv clones based on either DNA sequence or amino acid sequence, respectively, for the two % values shown separated by a/.

Analysis of CDR3 nucleotide length distributions shows that tT_reg and T_conv populations are distinct in adult donors

The CDR3 amino acid length distributions were compared between CD4 T cell subtypes in adult donors (see Supporting information, Figs S1 and S2 for all CDR3 a.a. length frequencies). Using the K–S test, we determined the largest difference (D) between tT_reg, overlap and T_conv distributions for the two adult donors (Supporting information, Figs S3 and S4). The P‐values calculated for all pairwise comparison of CDR3 length distributions were <0·0001, indicating that they are actually separate populations (alternative hypothesis) rather than separate samplings of the same population (null hypothesis).

Discussion

A limited number of prior studies have investigated the TCR repertoire of conventional, effector CD4 T cells compared to regulatory T cells in humans. Initially, prior to the ability to perform deep sequencing, it was first shown that the overall repertoires shared similar frequencies of Vβ genes 12. Spectratyping confirmed similar length distribution between T_conv and T_regs within Vβ genes 12. These studies suggested that there were no gross differences in the selection of which TCR gene segments undergo V(D)J rearrangement between the two functionally distinct cell types. A more recent study that compared CDR3 sequences directly in a limited number of only one or two TRVB genes found a not insignificant amount of overlap between fairly high‐frequency clonotypes and estimated the overall overlap between T_conv and T_reg repertoires to be 24% 7.

Based on our current data from deep sequencing of the entire TCR repertoire in both adult and cord blood donors, we believe that this prior finding was a major over‐estimate and that the actual overlap between T_conv and T_regs is less than 1%. Our conclusions are based on a more highly refined strategy for purifying and confirming bona fide tT_regs and they are also based on investigating both adult and neonatal T cells. The finding of virtual lack of overlap is also strengthened by the intentional inclusion of an overlap group, with which we found significant sharing of clones with both tT_reg and T_conv cells.

Analysis of TRBV and TRBJ pairing shows similar patterns for cord blood tT_regs and T_conv (Fig. 5). This suggests that the bulk of the differences in TCR repertoires between tT_regs and T_conv is due to junctional diversity, which has been shown previously to account for an approximate 5‐log fold increase in overall TCR repertoire diversity 13.

Studies using TCR transgenic mice that analysed sequences derived from tT_regs suggest that there is a limited tT_reg niche 14, 15, and simply expressing a tT_reg‐derived TCR does not determine CD4⁺ T cell developmental fate fully. Furthermore, expression of a gene programme favouring either the tT_reg or T_conv epigenetic landscape exists even prior to final lineage commitment and expression of master regulators such as FoxP3 3.

It has also been suggested that tT_regs bear a repertoire that favours recognition of self‐peptides in the periphery, whereas T_conv TCR favours foreign‐antigen recognition. However, our group (unpublished data) and others 16 have shown that tetramer major histocompatibility complex (MHC)II : antigen‐specific tT_regs and T_conv can recognize the same foreign antigen with distinct TCR CDR3 sequences.

Taken together, the aforementioned studies and the data reported in this paper support the concept that the tT_reg and T_conv develop separately in thymus, possibly being selected on distinct MHCII : peptide pairings. The observation of almost complete TCR repertoire separation suggests that unique antigen presentation regions in the thymus (different thymic location? different APCs? different local cytokine milieu?) or unique conditions for positive/negative selection favour distinct tT_reg and T_conv CD4⁺ T cell lineages. The data from comparison of adult T_reg and T_conv clones also suggest that there is very little, if any, conversion of T_conv into T_regs. So‐called peripherally induced T_regs, or pT_regs, have been well documented in mouse in‐vivo studies (summarized in reference [9]), but have yet to be demonstrated convincingly in humans.

Further research to identify the unique thymic compartments of tT_reg versus T_conv development will improve our understanding of tolerance. Furthermore, as has been observed in limited studies thus far 17, distinct monocyte‐derived dendritic cells may have the potential to stimulate preferentially and expand a pool of tT_regs (Helios⁺; TSDR demethylated), resulting in tolerance to a specific autoantigen without relying upon the conversion of antigen‐specific T_conv cells into pT_regs.

Disclosure

None.

Supporting information

Additional Supporting information may be found in the online version of this article at the publisher's web‐site:

Figs S1 and S2. Full CDR3 a.a. length distribution for all TRBV genes. The length distributions for each TRBV gene are shown for conventional T cell (T_conv), overlap and thymically derived regulatory T cell (tT_reg) groups for adult donor 118 (Fig. S1) and 210 (Fig. S2). The CDR3 length distribution ‘n’ refers to total reads so that it can reflect clones that are being selected for.

Figs S3 and S4. Statistical comparison of the CDR3 amino acid length distributions between CD4 T cell subtypes in adults. Using the K–S test, as described in the Methods section, we determined the largest difference (D) between thymically derived regulatory T cell (tT_reg), overlap and conventional T cell (T_conv) distributions for the two adult donors (shown, respectively, in Figs S3 and S4). The P‐values calculated for all pairwise comparison of CDR3 length distributions were < 0·0001.